1\section{Conclusion}2\label{sec:conclusion}3We asked under which conditions each way of passing information between model executions preserves what matters at the lowest measured cost,4and we answered with a single harness on commodity hardware, real cache tensors, measured bytes and synchronized clocks, released records for5every attempt, and a protocol frozen before the test split was touched. The bounded findings are these. (1) Lossless representations are6interchangeable: text, its zstd encoding, the native cache, the reloaded cache and the 8-bit packed cache give identical answers; the choice7among them is a pure cost question, and the cost ranking flips between the cold case (text wins: no build) and the reused case (cache wins:8no prefill), with a 1~GbE link and a 3B model already favouring moving an 8-bit cache over re-reading at two thousand tokens.9(2) Every lossy choice we measured is model- or regime-dependent: 4-bit per-token quantization is free for Llama-3.2-3B and fatal for10Qwen2.5-3B; attention-based eviction is excellent with the question and useless without it, and no eviction survives a code specification;11LLMLingua-2 drops exactly the identifiers private-fact questions ask for. (3) At a 512-token sender budget, prose summaries carried more than12structured evidence with provenance, in single handoffs, in eight-hop relays and in fan-in, because provenance is paid in the sender's tokens;13relays themselves were nearly lossless after the first hop and invented almost nothing, and metered repair bought a few points for kilobytes of14re-reading. (4) A same-family linear KV mapper that reconstructs most of the receiver's cache variance transfers no content, and produces the15most dangerous kind of failure---coherent abstention---while an invalid cache concatenation produced plausible answers; compatibility manifests16and negative controls are therefore part of any cache-transfer protocol, not an afterthought. (5) A router fitted on development data matched17the best static policy at half the amortized latency when questions were unknown and offered nothing when they were known.1819Open questions follow directly. The structured-evidence result should be re-run with budgets that separate provenance from values and with20senders large enough to fill them; the break-even map needs a faster receiver and a slower link to locate its crossing; the transfer ladder's21cross-family rung needs a trained mapper with generation and distillation terms, and its evaluation must keep the sender-alone and22receiver-alone references and an identity control beside it; and the fixed-point behaviour of structured relays suggests that the first23message, not the chain, is where compression research for multi-agent systems should concentrate.24